wombatsock

• An LLM hard-coded into silicon that can do inference at 17k tokens/s???

  Posted by wombatsock@reddit | LocalLLaMA | View on Reddit | 72 comments

• Inference on new Framework desktop

  Posted by wombatsock@reddit | LocalLLaMA | View on Reddit | 23 comments

• Renting your very own GPU from DigitalOcean

  Posted by wombatsock@reddit | LocalLLaMA | View on Reddit | 8 comments

• "Refusal in LLMs is mediated by a single direction" - research findings on a simple way to jailbreak any LLM

  Posted by wombatsock@reddit | LocalLLaMA | View on Reddit | 12 comments